G Gardian

Independent authority control for autonomous agents

Give AI agents autonomy. Keep the authority.

Gardian independently controls what autonomous agents are allowed to do, how much business risk they can create, when human approval is required, and what happens when authority must be revoked.

Watch Gardian stop an unauthorized action

Intent explains or constrains an action; intent does not authorize an action.

Live authority decision● PRESENTATIONAL DEMO
REQUEST$8,750 payment
→
AUTHORITYNOT GRANTED
→
DECISIONBLOCKED

Business-risk ceiling $10,000 ✓

External calls 0

BLOCKED BEFORE DISPATCHACTION_NOT_GRANTEDNothing left Gardian

Being under the financial limit did not create payment authority.

A legitimate-looking request can still be unauthorized

Access is not authority.

Identity✓
Mission✓
Intent✓
Tool access✓
Resource access✓
Risk ceiling✓
Business authority ✕
BLOCK — ACTION_NOT_GRANTED

Mission, intent, and tool access do not create delegated business authority.

Delegated business authority

Every delegation narrows authority.

Follow one supplier-payment branch from the Owner to the execution edge. Each child receives a smaller, explicit scope; none can expand itself.

Verified human approval

Approval for the exact action. In the exact moment.

Agent-generated text, bearer-token possession, or “the owner said yes” are not enough.

This visual explanation does not invoke the real protected Gardian app or WebAuthn.

Add supplier — Northstar Components

✓Authenticated Owner

✓Fresh passkey verification

✓Exact action + resource

✓Current authority epoch

Verified Human Approval

Revocation & containment

One revocation fences the entire supplier-payment branch.

The same authority lineage shown above is invalidated before queued work can dispatch.

Negotiation AgentACTIVE
Execution AgentACTIVE
Supplier paymentACTIVE
Queued actionsACTIVE
Prior approvalsACTIVE

Blast radius & causality

Reconstruct this decision, not a pile of logs.

Negotiation Agent→$25k delegated scope→$8,750 supplier payment→
Authority checkedPayment blockedExternal calls: 0Child grant fencedQueue invalidatedEvidence sealed

The public scenario illustrates causal modeling; it does not claim a live payment integration.

GARDIAN / INCIDENT RECORD

Gardian Incident Evidence

✓Authority lineage verified
✓Human approval verified
✓Decision integrity verified
✓Dispatch state verified
✓Revocation verified
✓Containment verified
✓Audit chain verified

Evidence, not assurances

Reconstruct the decision and every downstream effect.

A premium evidence package for incident response and internal review—without claiming certification or a guarantee.

Founding 20

Think your AI agent is safe?

Let Gardian try to break its authority model.

We’re selecting 20 teams operating consequential autonomous workflows for a private, founder-led authority assessment.

Who qualifiesTeams with an agent that can affect money, production, customers, sensitive data, or delegated workers.
What it takesA three-minute intake and one 45-minute working session. No production credentials or confidential prompts.
What you receiveA private authority map, adversarial control findings, containment analysis, and prioritized control plan.
How privacy worksYour assessment stays private. Benchmark use is separate, optional, and structurally anonymized.
✓ Agent Risk Score✓ Authority Map✓ Approval Analysis✓ Containment Assessment✓ Blast-Radius Analysis✓ Adversarial Control Test✓ Private Risk Report
Applications are reviewed for fit. No purchase required.

What’s the one action you don’t trust your AI agent to perform autonomously?